fs/exec.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/fs/exec.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  * #!-checking implemented by tytso.
  10  */
  11 /*
  12  * Demand-loading implemented 01.12.91 - no need to read anything but
  13  * the header into memory. The inode of the executable is put into
  14  * "current->executable", and page faults do the actual loading. Clean.
  15  *
  16  * Once more I can proudly say that linux stood up to being changed: it
  17  * was less than 2 hours work to get demand-loading completely implemented.
  18  *
  19  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  20  * current->executable is only used by the procfs.  This allows a dispatch
  21  * table to check for several different types  of binary formats.  We keep
  22  * trying until we recognize the file or we run out of supported binary
  23  * formats.
  24  */
  25
  26 #include <linux/slab.h>
  27 #include <linux/file.h>
  28 #include <linux/fdtable.h>
  29 #include <linux/mm.h>
  30 #include <linux/vmacache.h>
  31 #include <linux/stat.h>
  32 #include <linux/fcntl.h>
  33 #include <linux/swap.h>
  34 #include <linux/string.h>
  35 #include <linux/init.h>
  36 #include <linux/sched/mm.h>
  37 #include <linux/sched/coredump.h>
  38 #include <linux/sched/signal.h>
  39 #include <linux/sched/numa_balancing.h>
  40 #include <linux/sched/task.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/perf_event.h>
  43 #include <linux/highmem.h>
  44 #include <linux/spinlock.h>
  45 #include <linux/key.h>
  46 #include <linux/personality.h>
  47 #include <linux/binfmts.h>
  48 #include <linux/utsname.h>
  49 #include <linux/pid_namespace.h>
  50 #include <linux/module.h>
  51 #include <linux/namei.h>
  52 #include <linux/mount.h>
  53 #include <linux/security.h>
  54 #include <linux/syscalls.h>
  55 #include <linux/tsacct_kern.h>
  56 #include <linux/cn_proc.h>
  57 #include <linux/audit.h>
  58 #include <linux/tracehook.h>
  59 #include <linux/kmod.h>
  60 #include <linux/fsnotify.h>
  61 #include <linux/fs_struct.h>
  62 #include <linux/oom.h>
  63 #include <linux/compat.h>
  64 #include <linux/vmalloc.h>
  65
  66 #include <linux/uaccess.h>
  67 #include <asm/mmu_context.h>
  68 #include <asm/tlb.h>
  69
  70 #include <trace/events/task.h>
  71 #include "internal.h"
  72
  73 #include <trace/events/sched.h>
  74
  75 static int bprm_creds_from_file(struct linux_binprm *bprm);
  76
  77 int suid_dumpable = 0;
  78
  79 static LIST_HEAD(formats);
  80 static DEFINE_RWLOCK(binfmt_lock);
  81
  82 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  83 {
  84         BUG_ON(!fmt);
  85         if (WARN_ON(!fmt->load_binary))
  86                 return;
  87         write_lock(&binfmt_lock);
  88         insert ? list_add(&fmt->lh, &formats) :
  89                  list_add_tail(&fmt->lh, &formats);
  90         write_unlock(&binfmt_lock);
  91 }
  92
  93 EXPORT_SYMBOL(__register_binfmt);
  94
  95 void unregister_binfmt(struct linux_binfmt * fmt)
  96 {
  97         write_lock(&binfmt_lock);
  98         list_del(&fmt->lh);
  99         write_unlock(&binfmt_lock);
 100 }
 101
 102 EXPORT_SYMBOL(unregister_binfmt);
 103
 104 static inline void put_binfmt(struct linux_binfmt * fmt)
 105 {
 106         module_put(fmt->module);
 107 }
 108
 109 bool path_noexec(const struct path *path)
 110 {
 111         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 112                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 113 }
 114
 115 #ifdef CONFIG_USELIB
 116 /*
 117  * Note that a shared library must be both readable and executable due to
 118  * security reasons.
 119  *
 120  * Also note that we take the address to load from from the file itself.
 121  */
 122 SYSCALL_DEFINE1(uselib, const char __user *, library)
 123 {
 124         struct linux_binfmt *fmt;
 125         struct file *file;
 126         struct filename *tmp = getname(library);
 127         int error = PTR_ERR(tmp);
 128         static const struct open_flags uselib_flags = {
 129                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 130                 .acc_mode = MAY_READ | MAY_EXEC,
 131                 .intent = LOOKUP_OPEN,
 132                 .lookup_flags = LOOKUP_FOLLOW,
 133         };
 134
 135         if (IS_ERR(tmp))
 136                 goto out;
 137
 138         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 139         putname(tmp);
 140         error = PTR_ERR(file);
 141         if (IS_ERR(file))
 142                 goto out;
 143
 144         /*
 145          * may_open() has already checked for this, so it should be
 146          * impossible to trip now. But we need to be extra cautious
 147          * and check again at the very end too.
 148          */
 149         error = -EACCES;
 150         if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
 151                          path_noexec(&file->f_path)))
 152                 goto exit;
 153
 154         fsnotify_open(file);
 155
 156         error = -ENOEXEC;
 157
 158         read_lock(&binfmt_lock);
 159         list_for_each_entry(fmt, &formats, lh) {
 160                 if (!fmt->load_shlib)
 161                         continue;
 162                 if (!try_module_get(fmt->module))
 163                         continue;
 164                 read_unlock(&binfmt_lock);
 165                 error = fmt->load_shlib(file);
 166                 read_lock(&binfmt_lock);
 167                 put_binfmt(fmt);
 168                 if (error != -ENOEXEC)
 169                         break;
 170         }
 171         read_unlock(&binfmt_lock);
 172 exit:
 173         fput(file);
 174 out:
 175         return error;
 176 }
 177 #endif /* #ifdef CONFIG_USELIB */
 178
 179 #ifdef CONFIG_MMU
 180 /*
 181  * The nascent bprm->mm is not visible until exec_mmap() but it can
 182  * use a lot of memory, account these pages in current->mm temporary
 183  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 184  * change the counter back via acct_arg_size(0).
 185  */
 186 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 187 {
 188         struct mm_struct *mm = current->mm;
 189         long diff = (long)(pages - bprm->vma_pages);
 190
 191         if (!mm || !diff)
 192                 return;
 193
 194         bprm->vma_pages = pages;
 195         add_mm_counter(mm, MM_ANONPAGES, diff);
 196 }
 197
 198 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 199                 int write)
 200 {
 201         struct page *page;
 202         int ret;
 203         unsigned int gup_flags = FOLL_FORCE;
 204
 205 #ifdef CONFIG_STACK_GROWSUP
 206         if (write) {
 207                 ret = expand_downwards(bprm->vma, pos);
 208                 if (ret < 0)
 209                         return NULL;
 210         }
 211 #endif
 212
 213         if (write)
 214                 gup_flags |= FOLL_WRITE;
 215
 216         /*
 217          * We are doing an exec().  'current' is the process
 218          * doing the exec and bprm->mm is the new process's mm.
 219          */
 220         ret = get_user_pages_remote(bprm->mm, pos, 1, gup_flags,
 221                         &page, NULL, NULL);
 222         if (ret <= 0)
 223                 return NULL;
 224
 225         if (write)
 226                 acct_arg_size(bprm, vma_pages(bprm->vma));
 227
 228         return page;
 229 }
 230
 231 static void put_arg_page(struct page *page)
 232 {
 233         put_page(page);
 234 }
 235
 236 static void free_arg_pages(struct linux_binprm *bprm)
 237 {
 238 }
 239
 240 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 241                 struct page *page)
 242 {
 243         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 244 }
 245
 246 static int __bprm_mm_init(struct linux_binprm *bprm)
 247 {
 248         int err;
 249         struct vm_area_struct *vma = NULL;
 250         struct mm_struct *mm = bprm->mm;
 251
 252         bprm->vma = vma = vm_area_alloc(mm);
 253         if (!vma)
 254                 return -ENOMEM;
 255         vma_set_anonymous(vma);
 256
 257         if (mmap_write_lock_killable(mm)) {
 258                 err = -EINTR;
 259                 goto err_free;
 260         }
 261
 262         /*
 263          * Place the stack at the largest stack address the architecture
 264          * supports. Later, we'll move this to an appropriate place. We don't
 265          * use STACK_TOP because that can depend on attributes which aren't
 266          * configured yet.
 267          */
 268         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 269         vma->vm_end = STACK_TOP_MAX;
 270         vma->vm_start = vma->vm_end - PAGE_SIZE;
 271         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 272         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 273
 274         err = insert_vm_struct(mm, vma);
 275         if (err)
 276                 goto err;
 277
 278         mm->stack_vm = mm->total_vm = 1;
 279         mmap_write_unlock(mm);
 280         bprm->p = vma->vm_end - sizeof(void *);
 281         return 0;
 282 err:
 283         mmap_write_unlock(mm);
 284 err_free:
 285         bprm->vma = NULL;
 286         vm_area_free(vma);
 287         return err;
 288 }
 289
 290 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 291 {
 292         return len <= MAX_ARG_STRLEN;
 293 }
 294
 295 #else
 296
 297 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 298 {
 299 }
 300
 301 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 302                 int write)
 303 {
 304         struct page *page;
 305
 306         page = bprm->page[pos / PAGE_SIZE];
 307         if (!page && write) {
 308                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 309                 if (!page)
 310                         return NULL;
 311                 bprm->page[pos / PAGE_SIZE] = page;
 312         }
 313
 314         return page;
 315 }
 316
 317 static void put_arg_page(struct page *page)
 318 {
 319 }
 320
 321 static void free_arg_page(struct linux_binprm *bprm, int i)
 322 {
 323         if (bprm->page[i]) {
 324                 __free_page(bprm->page[i]);
 325                 bprm->page[i] = NULL;
 326         }
 327 }
 328
 329 static void free_arg_pages(struct linux_binprm *bprm)
 330 {
 331         int i;
 332
 333         for (i = 0; i < MAX_ARG_PAGES; i++)
 334                 free_arg_page(bprm, i);
 335 }
 336
 337 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 338                 struct page *page)
 339 {
 340 }
 341
 342 static int __bprm_mm_init(struct linux_binprm *bprm)
 343 {
 344         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 345         return 0;
 346 }
 347
 348 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 349 {
 350         return len <= bprm->p;
 351 }
 352
 353 #endif /* CONFIG_MMU */
 354
 355 /*
 356  * Create a new mm_struct and populate it with a temporary stack
 357  * vm_area_struct.  We don't have enough context at this point to set the stack
 358  * flags, permissions, and offset, so we use temporary values.  We'll update
 359  * them later in setup_arg_pages().
 360  */
 361 static int bprm_mm_init(struct linux_binprm *bprm)
 362 {
 363         int err;
 364         struct mm_struct *mm = NULL;
 365
 366         bprm->mm = mm = mm_alloc();
 367         err = -ENOMEM;
 368         if (!mm)
 369                 goto err;
 370
 371         /* Save current stack limit for all calculations made during exec. */
 372         task_lock(current->group_leader);
 373         bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
 374         task_unlock(current->group_leader);
 375
 376         err = __bprm_mm_init(bprm);
 377         if (err)
 378                 goto err;
 379
 380         return 0;
 381
 382 err:
 383         if (mm) {
 384                 bprm->mm = NULL;
 385                 mmdrop(mm);
 386         }
 387
 388         return err;
 389 }
 390
 391 struct user_arg_ptr {
 392 #ifdef CONFIG_COMPAT
 393         bool is_compat;
 394 #endif
 395         union {
 396                 const char __user *const __user *native;
 397 #ifdef CONFIG_COMPAT
 398                 const compat_uptr_t __user *compat;
 399 #endif
 400         } ptr;
 401 };
 402
 403 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 404 {
 405         const char __user *native;
 406
 407 #ifdef CONFIG_COMPAT
 408         if (unlikely(argv.is_compat)) {
 409                 compat_uptr_t compat;
 410
 411                 if (get_user(compat, argv.ptr.compat + nr))
 412                         return ERR_PTR(-EFAULT);
 413
 414                 return compat_ptr(compat);
 415         }
 416 #endif
 417
 418         if (get_user(native, argv.ptr.native + nr))
 419                 return ERR_PTR(-EFAULT);
 420
 421         return native;
 422 }
 423
 424 /*
 425  * count() counts the number of strings in array ARGV.
 426  */
 427 static int count(struct user_arg_ptr argv, int max)
 428 {
 429         int i = 0;
 430
 431         if (argv.ptr.native != NULL) {
 432                 for (;;) {
 433                         const char __user *p = get_user_arg_ptr(argv, i);
 434
 435                         if (!p)
 436                                 break;
 437
 438                         if (IS_ERR(p))
 439                                 return -EFAULT;
 440
 441                         if (i >= max)
 442                                 return -E2BIG;
 443                         ++i;
 444
 445                         if (fatal_signal_pending(current))
 446                                 return -ERESTARTNOHAND;
 447                         cond_resched();
 448                 }
 449         }
 450         return i;
 451 }
 452
 453 static int count_strings_kernel(const char *const *argv)
 454 {
 455         int i;
 456
 457         if (!argv)
 458                 return 0;
 459
 460         for (i = 0; argv[i]; ++i) {
 461                 if (i >= MAX_ARG_STRINGS)
 462                         return -E2BIG;
 463                 if (fatal_signal_pending(current))
 464                         return -ERESTARTNOHAND;
 465                 cond_resched();
 466         }
 467         return i;
 468 }
 469
 470 static int bprm_stack_limits(struct linux_binprm *bprm)
 471 {
 472         unsigned long limit, ptr_size;
 473
 474         /*
 475          * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
 476          * (whichever is smaller) for the argv+env strings.
 477          * This ensures that:
 478          *  - the remaining binfmt code will not run out of stack space,
 479          *  - the program will have a reasonable amount of stack left
 480          *    to work from.
 481          */
 482         limit = _STK_LIM / 4 * 3;
 483         limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
 484         /*
 485          * We've historically supported up to 32 pages (ARG_MAX)
 486          * of argument strings even with small stacks
 487          */
 488         limit = max_t(unsigned long, limit, ARG_MAX);
 489         /*
 490          * We must account for the size of all the argv and envp pointers to
 491          * the argv and envp strings, since they will also take up space in
 492          * the stack. They aren't stored until much later when we can't
 493          * signal to the parent that the child has run out of stack space.
 494          * Instead, calculate it here so it's possible to fail gracefully.
 495          */
 496         ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
 497         if (limit <= ptr_size)
 498                 return -E2BIG;
 499         limit -= ptr_size;
 500
 501         bprm->argmin = bprm->p - limit;
 502         return 0;
 503 }
 504
 505 /*
 506  * 'copy_strings()' copies argument/environment strings from the old
 507  * processes's memory to the new process's stack.  The call to get_user_pages()
 508  * ensures the destination page is created and not swapped out.
 509  */
 510 static int copy_strings(int argc, struct user_arg_ptr argv,
 511                         struct linux_binprm *bprm)
 512 {
 513         struct page *kmapped_page = NULL;
 514         char *kaddr = NULL;
 515         unsigned long kpos = 0;
 516         int ret;
 517
 518         while (argc-- > 0) {
 519                 const char __user *str;
 520                 int len;
 521                 unsigned long pos;
 522
 523                 ret = -EFAULT;
 524                 str = get_user_arg_ptr(argv, argc);
 525                 if (IS_ERR(str))
 526                         goto out;
 527
 528                 len = strnlen_user(str, MAX_ARG_STRLEN);
 529                 if (!len)
 530                         goto out;
 531
 532                 ret = -E2BIG;
 533                 if (!valid_arg_len(bprm, len))
 534                         goto out;
 535
 536                 /* We're going to work our way backwords. */
 537                 pos = bprm->p;
 538                 str += len;
 539                 bprm->p -= len;
 540 #ifdef CONFIG_MMU
 541                 if (bprm->p < bprm->argmin)
 542                         goto out;
 543 #endif
 544
 545                 while (len > 0) {
 546                         int offset, bytes_to_copy;
 547
 548                         if (fatal_signal_pending(current)) {
 549                                 ret = -ERESTARTNOHAND;
 550                                 goto out;
 551                         }
 552                         cond_resched();
 553
 554                         offset = pos % PAGE_SIZE;
 555                         if (offset == 0)
 556                                 offset = PAGE_SIZE;
 557
 558                         bytes_to_copy = offset;
 559                         if (bytes_to_copy > len)
 560                                 bytes_to_copy = len;
 561
 562                         offset -= bytes_to_copy;
 563                         pos -= bytes_to_copy;
 564                         str -= bytes_to_copy;
 565                         len -= bytes_to_copy;
 566
 567                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 568                                 struct page *page;
 569
 570                                 page = get_arg_page(bprm, pos, 1);
 571                                 if (!page) {
 572                                         ret = -E2BIG;
 573                                         goto out;
 574                                 }
 575
 576                                 if (kmapped_page) {
 577                                         flush_kernel_dcache_page(kmapped_page);
 578                                         kunmap(kmapped_page);
 579                                         put_arg_page(kmapped_page);
 580                                 }
 581                                 kmapped_page = page;
 582                                 kaddr = kmap(kmapped_page);
 583                                 kpos = pos & PAGE_MASK;
 584                                 flush_arg_page(bprm, kpos, kmapped_page);
 585                         }
 586                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 587                                 ret = -EFAULT;
 588                                 goto out;
 589                         }
 590                 }
 591         }
 592         ret = 0;
 593 out:
 594         if (kmapped_page) {
 595                 flush_kernel_dcache_page(kmapped_page);
 596                 kunmap(kmapped_page);
 597                 put_arg_page(kmapped_page);
 598         }
 599         return ret;
 600 }
 601
 602 /*
 603  * Copy and argument/environment string from the kernel to the processes stack.
 604  */
 605 int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
 606 {
 607         int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
 608         unsigned long pos = bprm->p;
 609
 610         if (len == 0)
 611                 return -EFAULT;
 612         if (!valid_arg_len(bprm, len))
 613                 return -E2BIG;
 614
 615         /* We're going to work our way backwards. */
 616         arg += len;
 617         bprm->p -= len;
 618         if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
 619                 return -E2BIG;
 620
 621         while (len > 0) {
 622                 unsigned int bytes_to_copy = min_t(unsigned int, len,
 623                                 min_not_zero(offset_in_page(pos), PAGE_SIZE));
 624                 struct page *page;
 625                 char *kaddr;
 626
 627                 pos -= bytes_to_copy;
 628                 arg -= bytes_to_copy;
 629                 len -= bytes_to_copy;
 630
 631                 page = get_arg_page(bprm, pos, 1);
 632                 if (!page)
 633                         return -E2BIG;
 634                 kaddr = kmap_atomic(page);
 635                 flush_arg_page(bprm, pos & PAGE_MASK, page);
 636                 memcpy(kaddr + offset_in_page(pos), arg, bytes_to_copy);
 637                 flush_kernel_dcache_page(page);
 638                 kunmap_atomic(kaddr);
 639                 put_arg_page(page);
 640         }
 641
 642         return 0;
 643 }
 644 EXPORT_SYMBOL(copy_string_kernel);
 645
 646 static int copy_strings_kernel(int argc, const char *const *argv,
 647                                struct linux_binprm *bprm)
 648 {
 649         while (argc-- > 0) {
 650                 int ret = copy_string_kernel(argv[argc], bprm);
 651                 if (ret < 0)
 652                         return ret;
 653                 if (fatal_signal_pending(current))
 654                         return -ERESTARTNOHAND;
 655                 cond_resched();
 656         }
 657         return 0;
 658 }
 659
 660 #ifdef CONFIG_MMU
 661
 662 /*
 663  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 664  * the binfmt code determines where the new stack should reside, we shift it to
 665  * its final location.  The process proceeds as follows:
 666  *
 667  * 1) Use shift to calculate the new vma endpoints.
 668  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 669  *    arguments passed to subsequent functions are consistent.
 670  * 3) Move vma's page tables to the new range.
 671  * 4) Free up any cleared pgd range.
 672  * 5) Shrink the vma to cover only the new range.
 673  */
 674 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 675 {
 676         struct mm_struct *mm = vma->vm_mm;
 677         unsigned long old_start = vma->vm_start;
 678         unsigned long old_end = vma->vm_end;
 679         unsigned long length = old_end - old_start;
 680         unsigned long new_start = old_start - shift;
 681         unsigned long new_end = old_end - shift;
 682         struct mmu_gather tlb;
 683
 684         BUG_ON(new_start > new_end);
 685
 686         /*
 687          * ensure there are no vmas between where we want to go
 688          * and where we are
 689          */
 690         if (vma != find_vma(mm, new_start))
 691                 return -EFAULT;
 692
 693         /*
 694          * cover the whole range: [new_start, old_end)
 695          */
 696         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 697                 return -ENOMEM;
 698
 699         /*
 700          * move the page tables downwards, on failure we rely on
 701          * process cleanup to remove whatever mess we made.
 702          */
 703         if (length != move_page_tables(vma, old_start,
 704                                        vma, new_start, length, false))
 705                 return -ENOMEM;
 706
 707         lru_add_drain();
 708         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 709         if (new_end > old_start) {
 710                 /*
 711                  * when the old and new regions overlap clear from new_end.
 712                  */
 713                 free_pgd_range(&tlb, new_end, old_end, new_end,
 714                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 715         } else {
 716                 /*
 717                  * otherwise, clean from old_start; this is done to not touch
 718                  * the address space in [new_end, old_start) some architectures
 719                  * have constraints on va-space that make this illegal (IA64) -
 720                  * for the others its just a little faster.
 721                  */
 722                 free_pgd_range(&tlb, old_start, old_end, new_end,
 723                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 724         }
 725         tlb_finish_mmu(&tlb, old_start, old_end);
 726
 727         /*
 728          * Shrink the vma to just the new range.  Always succeeds.
 729          */
 730         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 731
 732         return 0;
 733 }
 734
 735 /*
 736  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 737  * the stack is optionally relocated, and some extra space is added.
 738  */
 739 int setup_arg_pages(struct linux_binprm *bprm,
 740                     unsigned long stack_top,
 741                     int executable_stack)
 742 {
 743         unsigned long ret;
 744         unsigned long stack_shift;
 745         struct mm_struct *mm = current->mm;
 746         struct vm_area_struct *vma = bprm->vma;
 747         struct vm_area_struct *prev = NULL;
 748         unsigned long vm_flags;
 749         unsigned long stack_base;
 750         unsigned long stack_size;
 751         unsigned long stack_expand;
 752         unsigned long rlim_stack;
 753
 754 #ifdef CONFIG_STACK_GROWSUP
 755         /* Limit stack size */
 756         stack_base = bprm->rlim_stack.rlim_max;
 757         if (stack_base > STACK_SIZE_MAX)
 758                 stack_base = STACK_SIZE_MAX;
 759
 760         /* Add space for stack randomization. */
 761         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 762
 763         /* Make sure we didn't let the argument array grow too large. */
 764         if (vma->vm_end - vma->vm_start > stack_base)
 765                 return -ENOMEM;
 766
 767         stack_base = PAGE_ALIGN(stack_top - stack_base);
 768
 769         stack_shift = vma->vm_start - stack_base;
 770         mm->arg_start = bprm->p - stack_shift;
 771         bprm->p = vma->vm_end - stack_shift;
 772 #else
 773         stack_top = arch_align_stack(stack_top);
 774         stack_top = PAGE_ALIGN(stack_top);
 775
 776         if (unlikely(stack_top < mmap_min_addr) ||
 777             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 778                 return -ENOMEM;
 779
 780         stack_shift = vma->vm_end - stack_top;
 781
 782         bprm->p -= stack_shift;
 783         mm->arg_start = bprm->p;
 784 #endif
 785
 786         if (bprm->loader)
 787                 bprm->loader -= stack_shift;
 788         bprm->exec -= stack_shift;
 789
 790         if (mmap_write_lock_killable(mm))
 791                 return -EINTR;
 792
 793         vm_flags = VM_STACK_FLAGS;
 794
 795         /*
 796          * Adjust stack execute permissions; explicitly enable for
 797          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 798          * (arch default) otherwise.
 799          */
 800         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 801                 vm_flags |= VM_EXEC;
 802         else if (executable_stack == EXSTACK_DISABLE_X)
 803                 vm_flags &= ~VM_EXEC;
 804         vm_flags |= mm->def_flags;
 805         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 806
 807         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 808                         vm_flags);
 809         if (ret)
 810                 goto out_unlock;
 811         BUG_ON(prev != vma);
 812
 813         if (unlikely(vm_flags & VM_EXEC)) {
 814                 pr_warn_once("process '%pD4' started with executable stack\n",
 815                              bprm->file);
 816         }
 817
 818         /* Move stack pages down in memory. */
 819         if (stack_shift) {
 820                 ret = shift_arg_pages(vma, stack_shift);
 821                 if (ret)
 822                         goto out_unlock;
 823         }
 824
 825         /* mprotect_fixup is overkill to remove the temporary stack flags */
 826         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 827
 828         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 829         stack_size = vma->vm_end - vma->vm_start;
 830         /*
 831          * Align this down to a page boundary as expand_stack
 832          * will align it up.
 833          */
 834         rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
 835 #ifdef CONFIG_STACK_GROWSUP
 836         if (stack_size + stack_expand > rlim_stack)
 837                 stack_base = vma->vm_start + rlim_stack;
 838         else
 839                 stack_base = vma->vm_end + stack_expand;
 840 #else
 841         if (stack_size + stack_expand > rlim_stack)
 842                 stack_base = vma->vm_end - rlim_stack;
 843         else
 844                 stack_base = vma->vm_start - stack_expand;
 845 #endif
 846         current->mm->start_stack = bprm->p;
 847         ret = expand_stack(vma, stack_base);
 848         if (ret)
 849                 ret = -EFAULT;
 850
 851 out_unlock:
 852         mmap_write_unlock(mm);
 853         return ret;
 854 }
 855 EXPORT_SYMBOL(setup_arg_pages);
 856
 857 #else
 858
 859 /*
 860  * Transfer the program arguments and environment from the holding pages
 861  * onto the stack. The provided stack pointer is adjusted accordingly.
 862  */
 863 int transfer_args_to_stack(struct linux_binprm *bprm,
 864                            unsigned long *sp_location)
 865 {
 866         unsigned long index, stop, sp;
 867         int ret = 0;
 868
 869         stop = bprm->p >> PAGE_SHIFT;
 870         sp = *sp_location;
 871
 872         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 873                 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 874                 char *src = kmap(bprm->page[index]) + offset;
 875                 sp -= PAGE_SIZE - offset;
 876                 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 877                         ret = -EFAULT;
 878                 kunmap(bprm->page[index]);
 879                 if (ret)
 880                         goto out;
 881         }
 882
 883         *sp_location = sp;
 884
 885 out:
 886         return ret;
 887 }
 888 EXPORT_SYMBOL(transfer_args_to_stack);
 889
 890 #endif /* CONFIG_MMU */
 891
 892 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 893 {
 894         struct file *file;
 895         int err;
 896         struct open_flags open_exec_flags = {
 897                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 898                 .acc_mode = MAY_EXEC,
 899                 .intent = LOOKUP_OPEN,
 900                 .lookup_flags = LOOKUP_FOLLOW,
 901         };
 902
 903         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 904                 return ERR_PTR(-EINVAL);
 905         if (flags & AT_SYMLINK_NOFOLLOW)
 906                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 907         if (flags & AT_EMPTY_PATH)
 908                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 909
 910         file = do_filp_open(fd, name, &open_exec_flags);
 911         if (IS_ERR(file))
 912                 goto out;
 913
 914         /*
 915          * may_open() has already checked for this, so it should be
 916          * impossible to trip now. But we need to be extra cautious
 917          * and check again at the very end too.
 918          */
 919         err = -EACCES;
 920         if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
 921                          path_noexec(&file->f_path)))
 922                 goto exit;
 923
 924         err = deny_write_access(file);
 925         if (err)
 926                 goto exit;
 927
 928         if (name->name[0] != '\0')
 929                 fsnotify_open(file);
 930
 931 out:
 932         return file;
 933
 934 exit:
 935         fput(file);
 936         return ERR_PTR(err);
 937 }
 938
 939 struct file *open_exec(const char *name)
 940 {
 941         struct filename *filename = getname_kernel(name);
 942         struct file *f = ERR_CAST(filename);
 943
 944         if (!IS_ERR(filename)) {
 945                 f = do_open_execat(AT_FDCWD, filename, 0);
 946                 putname(filename);
 947         }
 948         return f;
 949 }
 950 EXPORT_SYMBOL(open_exec);
 951
 952 int kernel_read_file(struct file *file, void **buf, loff_t *size,
 953                      loff_t max_size, enum kernel_read_file_id id)
 954 {
 955         loff_t i_size, pos;
 956         ssize_t bytes = 0;
 957         void *allocated = NULL;
 958         int ret;
 959
 960         if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
 961                 return -EINVAL;
 962
 963         ret = deny_write_access(file);
 964         if (ret)
 965                 return ret;
 966
 967         ret = security_kernel_read_file(file, id);
 968         if (ret)
 969                 goto out;
 970
 971         i_size = i_size_read(file_inode(file));
 972         if (i_size <= 0) {
 973                 ret = -EINVAL;
 974                 goto out;
 975         }
 976         if (i_size > SIZE_MAX || (max_size > 0 && i_size > max_size)) {
 977                 ret = -EFBIG;
 978                 goto out;
 979         }
 980
 981         if (!*buf)
 982                 *buf = allocated = vmalloc(i_size);
 983         if (!*buf) {
 984                 ret = -ENOMEM;
 985                 goto out;
 986         }
 987
 988         pos = 0;
 989         while (pos < i_size) {
 990                 bytes = kernel_read(file, *buf + pos, i_size - pos, &pos);
 991                 if (bytes < 0) {
 992                         ret = bytes;
 993                         goto out_free;
 994                 }
 995
 996                 if (bytes == 0)
 997                         break;
 998         }
 999
1000         if (pos != i_size) {
1001                 ret = -EIO;
1002                 goto out_free;
1003         }
1004
1005         ret = security_kernel_post_read_file(file, *buf, i_size, id);
1006         if (!ret)
1007                 *size = pos;
1008
1009 out_free:
1010         if (ret < 0) {
1011                 if (allocated) {
1012                         vfree(*buf);
1013                         *buf = NULL;
1014                 }
1015         }
1016
1017 out:
1018         allow_write_access(file);
1019         return ret;
1020 }
1021 EXPORT_SYMBOL_GPL(kernel_read_file);
1022
1023 int kernel_read_file_from_path(const char *path, void **buf, loff_t *size,
1024                                loff_t max_size, enum kernel_read_file_id id)
1025 {
1026         struct file *file;
1027         int ret;
1028
1029         if (!path || !*path)
1030                 return -EINVAL;
1031
1032         file = filp_open(path, O_RDONLY, 0);
1033         if (IS_ERR(file))
1034                 return PTR_ERR(file);
1035
1036         ret = kernel_read_file(file, buf, size, max_size, id);
1037         fput(file);
1038         return ret;
1039 }
1040 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
1041
1042 int kernel_read_file_from_path_initns(const char *path, void **buf,
1043                                       loff_t *size, loff_t max_size,
1044                                       enum kernel_read_file_id id)
1045 {
1046         struct file *file;
1047         struct path root;
1048         int ret;
1049
1050         if (!path || !*path)
1051                 return -EINVAL;
1052
1053         task_lock(&init_task);
1054         get_fs_root(init_task.fs, &root);
1055         task_unlock(&init_task);
1056
1057         file = file_open_root(root.dentry, root.mnt, path, O_RDONLY, 0);
1058         path_put(&root);
1059         if (IS_ERR(file))
1060                 return PTR_ERR(file);
1061
1062         ret = kernel_read_file(file, buf, size, max_size, id);
1063         fput(file);
1064         return ret;
1065 }
1066 EXPORT_SYMBOL_GPL(kernel_read_file_from_path_initns);
1067
1068 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
1069                              enum kernel_read_file_id id)
1070 {
1071         struct fd f = fdget(fd);
1072         int ret = -EBADF;
1073
1074         if (!f.file)
1075                 goto out;
1076
1077         ret = kernel_read_file(f.file, buf, size, max_size, id);
1078 out:
1079         fdput(f);
1080         return ret;
1081 }
1082 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
1083
1084 #if defined(CONFIG_HAVE_AOUT) || defined(CONFIG_BINFMT_FLAT) || \
1085     defined(CONFIG_BINFMT_ELF_FDPIC)
1086 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
1087 {
1088         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
1089         if (res > 0)
1090                 flush_icache_user_range(addr, addr + len);
1091         return res;
1092 }
1093 EXPORT_SYMBOL(read_code);
1094 #endif
1095
1096 /*
1097  * Maps the mm_struct mm into the current task struct.
1098  * On success, this function returns with the mutex
1099  * exec_update_mutex locked.
1100  */
1101 static int exec_mmap(struct mm_struct *mm)
1102 {
1103         struct task_struct *tsk;
1104         struct mm_struct *old_mm, *active_mm;
1105         int ret;
1106
1107         /* Notify parent that we're no longer interested in the old VM */
1108         tsk = current;
1109         old_mm = current->mm;
1110         exec_mm_release(tsk, old_mm);
1111         if (old_mm)
1112                 sync_mm_rss(old_mm);
1113
1114         ret = mutex_lock_killable(&tsk->signal->exec_update_mutex);
1115         if (ret)
1116                 return ret;
1117
1118         if (old_mm) {
1119                 /*
1120                  * Make sure that if there is a core dump in progress
1121                  * for the old mm, we get out and die instead of going
1122                  * through with the exec.  We must hold mmap_lock around
1123                  * checking core_state and changing tsk->mm.
1124                  */
1125                 mmap_read_lock(old_mm);
1126                 if (unlikely(old_mm->core_state)) {
1127                         mmap_read_unlock(old_mm);
1128                         mutex_unlock(&tsk->signal->exec_update_mutex);
1129                         return -EINTR;
1130                 }
1131         }
1132
1133         task_lock(tsk);
1134         active_mm = tsk->active_mm;
1135         membarrier_exec_mmap(mm);
1136         tsk->mm = mm;
1137         tsk->active_mm = mm;
1138         activate_mm(active_mm, mm);
1139         tsk->mm->vmacache_seqnum = 0;
1140         vmacache_flush(tsk);
1141         task_unlock(tsk);
1142         if (old_mm) {
1143                 mmap_read_unlock(old_mm);
1144                 BUG_ON(active_mm != old_mm);
1145                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1146                 mm_update_next_owner(old_mm);
1147                 mmput(old_mm);
1148                 return 0;
1149         }
1150         mmdrop(active_mm);
1151         return 0;
1152 }
1153
1154 static int de_thread(struct task_struct *tsk)
1155 {
1156         struct signal_struct *sig = tsk->signal;
1157         struct sighand_struct *oldsighand = tsk->sighand;
1158         spinlock_t *lock = &oldsighand->siglock;
1159
1160         if (thread_group_empty(tsk))
1161                 goto no_thread_group;
1162
1163         /*
1164          * Kill all other threads in the thread group.
1165          */
1166         spin_lock_irq(lock);
1167         if (signal_group_exit(sig)) {
1168                 /*
1169                  * Another group action in progress, just
1170                  * return so that the signal is processed.
1171                  */
1172                 spin_unlock_irq(lock);
1173                 return -EAGAIN;
1174         }
1175
1176         sig->group_exit_task = tsk;
1177         sig->notify_count = zap_other_threads(tsk);
1178         if (!thread_group_leader(tsk))
1179                 sig->notify_count--;
1180
1181         while (sig->notify_count) {
1182                 __set_current_state(TASK_KILLABLE);
1183                 spin_unlock_irq(lock);
1184                 schedule();
1185                 if (__fatal_signal_pending(tsk))
1186                         goto killed;
1187                 spin_lock_irq(lock);
1188         }
1189         spin_unlock_irq(lock);
1190
1191         /*
1192          * At this point all other threads have exited, all we have to
1193          * do is to wait for the thread group leader to become inactive,
1194          * and to assume its PID:
1195          */
1196         if (!thread_group_leader(tsk)) {
1197                 struct task_struct *leader = tsk->group_leader;
1198
1199                 for (;;) {
1200                         cgroup_threadgroup_change_begin(tsk);
1201                         write_lock_irq(&tasklist_lock);
1202                         /*
1203                          * Do this under tasklist_lock to ensure that
1204                          * exit_notify() can't miss ->group_exit_task
1205                          */
1206                         sig->notify_count = -1;
1207                         if (likely(leader->exit_state))
1208                                 break;
1209                         __set_current_state(TASK_KILLABLE);
1210                         write_unlock_irq(&tasklist_lock);
1211                         cgroup_threadgroup_change_end(tsk);
1212                         schedule();
1213                         if (__fatal_signal_pending(tsk))
1214                                 goto killed;
1215                 }
1216
1217                 /*
1218                  * The only record we have of the real-time age of a
1219                  * process, regardless of execs it's done, is start_time.
1220                  * All the past CPU time is accumulated in signal_struct
1221                  * from sister threads now dead.  But in this non-leader
1222                  * exec, nothing survives from the original leader thread,
1223                  * whose birth marks the true age of this process now.
1224                  * When we take on its identity by switching to its PID, we
1225                  * also take its birthdate (always earlier than our own).
1226                  */
1227                 tsk->start_time = leader->start_time;
1228                 tsk->start_boottime = leader->start_boottime;
1229
1230                 BUG_ON(!same_thread_group(leader, tsk));
1231                 /*
1232                  * An exec() starts a new thread group with the
1233                  * TGID of the previous thread group. Rehash the
1234                  * two threads with a switched PID, and release
1235                  * the former thread group leader:
1236                  */
1237
1238                 /* Become a process group leader with the old leader's pid.
1239                  * The old leader becomes a thread of the this thread group.
1240                  */
1241                 exchange_tids(tsk, leader);
1242                 transfer_pid(leader, tsk, PIDTYPE_TGID);
1243                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1244                 transfer_pid(leader, tsk, PIDTYPE_SID);
1245
1246                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1247                 list_replace_init(&leader->sibling, &tsk->sibling);
1248
1249                 tsk->group_leader = tsk;
1250                 leader->group_leader = tsk;
1251
1252                 tsk->exit_signal = SIGCHLD;
1253                 leader->exit_signal = -1;
1254
1255                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1256                 leader->exit_state = EXIT_DEAD;
1257
1258                 /*
1259                  * We are going to release_task()->ptrace_unlink() silently,
1260                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1261                  * the tracer wont't block again waiting for this thread.
1262                  */
1263                 if (unlikely(leader->ptrace))
1264                         __wake_up_parent(leader, leader->parent);
1265                 write_unlock_irq(&tasklist_lock);
1266                 cgroup_threadgroup_change_end(tsk);
1267
1268                 release_task(leader);
1269         }
1270
1271         sig->group_exit_task = NULL;
1272         sig->notify_count = 0;
1273
1274 no_thread_group:
1275         /* we have changed execution domain */
1276         tsk->exit_signal = SIGCHLD;
1277
1278         BUG_ON(!thread_group_leader(tsk));
1279         return 0;
1280
1281 killed:
1282         /* protects against exit_notify() and __exit_signal() */
1283         read_lock(&tasklist_lock);
1284         sig->group_exit_task = NULL;
1285         sig->notify_count = 0;
1286         read_unlock(&tasklist_lock);
1287         return -EAGAIN;
1288 }
1289
1290
1291 /*
1292  * This function makes sure the current process has its own signal table,
1293  * so that flush_signal_handlers can later reset the handlers without
1294  * disturbing other processes.  (Other processes might share the signal
1295  * table via the CLONE_SIGHAND option to clone().)
1296  */
1297 static int unshare_sighand(struct task_struct *me)
1298 {
1299         struct sighand_struct *oldsighand = me->sighand;
1300
1301         if (refcount_read(&oldsighand->count) != 1) {
1302                 struct sighand_struct *newsighand;
1303                 /*
1304                  * This ->sighand is shared with the CLONE_SIGHAND
1305                  * but not CLONE_THREAD task, switch to the new one.
1306                  */
1307                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1308                 if (!newsighand)
1309                         return -ENOMEM;
1310
1311                 refcount_set(&newsighand->count, 1);
1312                 memcpy(newsighand->action, oldsighand->action,
1313                        sizeof(newsighand->action));
1314
1315                 write_lock_irq(&tasklist_lock);
1316                 spin_lock(&oldsighand->siglock);
1317                 rcu_assign_pointer(me->sighand, newsighand);
1318                 spin_unlock(&oldsighand->siglock);
1319                 write_unlock_irq(&tasklist_lock);
1320
1321                 __cleanup_sighand(oldsighand);
1322         }
1323         return 0;
1324 }
1325
1326 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1327 {
1328         task_lock(tsk);
1329         strncpy(buf, tsk->comm, buf_size);
1330         task_unlock(tsk);
1331         return buf;
1332 }
1333 EXPORT_SYMBOL_GPL(__get_task_comm);
1334
1335 /*
1336  * These functions flushes out all traces of the currently running executable
1337  * so that a new one can be started
1338  */
1339
1340 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1341 {
1342         task_lock(tsk);
1343         trace_task_rename(tsk, buf);
1344         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1345         task_unlock(tsk);
1346         perf_event_comm(tsk, exec);
1347 }
1348
1349 /*
1350  * Calling this is the point of no return. None of the failures will be
1351  * seen by userspace since either the process is already taking a fatal
1352  * signal (via de_thread() or coredump), or will have SEGV raised
1353  * (after exec_mmap()) by search_binary_handler (see below).
1354  */
1355 int begin_new_exec(struct linux_binprm * bprm)
1356 {
1357         struct task_struct *me = current;
1358         int retval;
1359
1360         /* Once we are committed compute the creds */
1361         retval = bprm_creds_from_file(bprm);
1362         if (retval)
1363                 return retval;
1364
1365         /*
1366          * Ensure all future errors are fatal.
1367          */
1368         bprm->point_of_no_return = true;
1369
1370         /*
1371          * Make this the only thread in the thread group.
1372          */
1373         retval = de_thread(me);
1374         if (retval)
1375                 goto out;
1376
1377         /*
1378          * Must be called _before_ exec_mmap() as bprm->mm is
1379          * not visibile until then. This also enables the update
1380          * to be lockless.
1381          */
1382         set_mm_exe_file(bprm->mm, bprm->file);
1383
1384         /* If the binary is not readable then enforce mm->dumpable=0 */
1385         would_dump(bprm, bprm->file);
1386         if (bprm->have_execfd)
1387                 would_dump(bprm, bprm->executable);
1388
1389         /*
1390          * Release all of the old mmap stuff
1391          */
1392         acct_arg_size(bprm, 0);
1393         retval = exec_mmap(bprm->mm);
1394         if (retval)
1395                 goto out;
1396
1397         bprm->mm = NULL;
1398
1399 #ifdef CONFIG_POSIX_TIMERS
1400         exit_itimers(me->signal);
1401         flush_itimer_signals();
1402 #endif
1403
1404         /*
1405          * Make the signal table private.
1406          */
1407         retval = unshare_sighand(me);
1408         if (retval)
1409                 goto out_unlock;
1410
1411         /*
1412          * Ensure that the uaccess routines can actually operate on userspace
1413          * pointers:
1414          */
1415         force_uaccess_begin();
1416
1417         me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1418                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1419         flush_thread();
1420         me->personality &= ~bprm->per_clear;
1421
1422         /*
1423          * We have to apply CLOEXEC before we change whether the process is
1424          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1425          * trying to access the should-be-closed file descriptors of a process
1426          * undergoing exec(2).
1427          */
1428         do_close_on_exec(me->files);
1429
1430         if (bprm->secureexec) {
1431                 /* Make sure parent cannot signal privileged process. */
1432                 me->pdeath_signal = 0;
1433
1434                 /*
1435                  * For secureexec, reset the stack limit to sane default to
1436                  * avoid bad behavior from the prior rlimits. This has to
1437                  * happen before arch_pick_mmap_layout(), which examines
1438                  * RLIMIT_STACK, but after the point of no return to avoid
1439                  * needing to clean up the change on failure.
1440                  */
1441                 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1442                         bprm->rlim_stack.rlim_cur = _STK_LIM;
1443         }
1444
1445         me->sas_ss_sp = me->sas_ss_size = 0;
1446
1447         /*
1448          * Figure out dumpability. Note that this checking only of current
1449          * is wrong, but userspace depends on it. This should be testing
1450          * bprm->secureexec instead.
1451          */
1452         if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1453             !(uid_eq(current_euid(), current_uid()) &&
1454               gid_eq(current_egid(), current_gid())))
1455                 set_dumpable(current->mm, suid_dumpable);
1456         else
1457                 set_dumpable(current->mm, SUID_DUMP_USER);
1458
1459         perf_event_exec();
1460         __set_task_comm(me, kbasename(bprm->filename), true);
1461
1462         /* An exec changes our domain. We are no longer part of the thread
1463            group */
1464         WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1465         flush_signal_handlers(me, 0);
1466
1467         /*
1468          * install the new credentials for this executable
1469          */
1470         security_bprm_committing_creds(bprm);
1471
1472         commit_creds(bprm->cred);
1473         bprm->cred = NULL;
1474
1475         /*
1476          * Disable monitoring for regular users
1477          * when executing setuid binaries. Must
1478          * wait until new credentials are committed
1479          * by commit_creds() above
1480          */
1481         if (get_dumpable(me->mm) != SUID_DUMP_USER)
1482                 perf_event_exit_task(me);
1483         /*
1484          * cred_guard_mutex must be held at least to this point to prevent
1485          * ptrace_attach() from altering our determination of the task's
1486          * credentials; any time after this it may be unlocked.
1487          */
1488         security_bprm_committed_creds(bprm);
1489
1490         /* Pass the opened binary to the interpreter. */
1491         if (bprm->have_execfd) {
1492                 retval = get_unused_fd_flags(0);
1493                 if (retval < 0)
1494                         goto out_unlock;
1495                 fd_install(retval, bprm->executable);
1496                 bprm->executable = NULL;
1497                 bprm->execfd = retval;
1498         }
1499         return 0;
1500
1501 out_unlock:
1502         mutex_unlock(&me->signal->exec_update_mutex);
1503 out:
1504         return retval;
1505 }
1506 EXPORT_SYMBOL(begin_new_exec);
1507
1508 void would_dump(struct linux_binprm *bprm, struct file *file)
1509 {
1510         struct inode *inode = file_inode(file);
1511         if (inode_permission(inode, MAY_READ) < 0) {
1512                 struct user_namespace *old, *user_ns;
1513                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1514
1515                 /* Ensure mm->user_ns contains the executable */
1516                 user_ns = old = bprm->mm->user_ns;
1517                 while ((user_ns != &init_user_ns) &&
1518                        !privileged_wrt_inode_uidgid(user_ns, inode))
1519                         user_ns = user_ns->parent;
1520
1521                 if (old != user_ns) {
1522                         bprm->mm->user_ns = get_user_ns(user_ns);
1523                         put_user_ns(old);
1524                 }
1525         }
1526 }
1527 EXPORT_SYMBOL(would_dump);
1528
1529 void setup_new_exec(struct linux_binprm * bprm)
1530 {
1531         /* Setup things that can depend upon the personality */
1532         struct task_struct *me = current;
1533
1534         arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1535
1536         arch_setup_new_exec();
1537
1538         /* Set the new mm task size. We have to do that late because it may
1539          * depend on TIF_32BIT which is only updated in flush_thread() on
1540          * some architectures like powerpc
1541          */
1542         me->mm->task_size = TASK_SIZE;
1543         mutex_unlock(&me->signal->exec_update_mutex);
1544         mutex_unlock(&me->signal->cred_guard_mutex);
1545 }
1546 EXPORT_SYMBOL(setup_new_exec);
1547
1548 /* Runs immediately before start_thread() takes over. */
1549 void finalize_exec(struct linux_binprm *bprm)
1550 {
1551         /* Store any stack rlimit changes before starting thread. */
1552         task_lock(current->group_leader);
1553         current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1554         task_unlock(current->group_leader);
1555 }
1556 EXPORT_SYMBOL(finalize_exec);
1557
1558 /*
1559  * Prepare credentials and lock ->cred_guard_mutex.
1560  * setup_new_exec() commits the new creds and drops the lock.
1561  * Or, if exec fails before, free_bprm() should release ->cred and
1562  * and unlock.
1563  */
1564 static int prepare_bprm_creds(struct linux_binprm *bprm)
1565 {
1566         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1567                 return -ERESTARTNOINTR;
1568
1569         bprm->cred = prepare_exec_creds();
1570         if (likely(bprm->cred))
1571                 return 0;
1572
1573         mutex_unlock(&current->signal->cred_guard_mutex);
1574         return -ENOMEM;
1575 }
1576
1577 static void free_bprm(struct linux_binprm *bprm)
1578 {
1579         if (bprm->mm) {
1580                 acct_arg_size(bprm, 0);
1581                 mmput(bprm->mm);
1582         }
1583         free_arg_pages(bprm);
1584         if (bprm->cred) {
1585                 mutex_unlock(&current->signal->cred_guard_mutex);
1586                 abort_creds(bprm->cred);
1587         }
1588         if (bprm->file) {
1589                 allow_write_access(bprm->file);
1590                 fput(bprm->file);
1591         }
1592         if (bprm->executable)
1593                 fput(bprm->executable);
1594         /* If a binfmt changed the interp, free it. */
1595         if (bprm->interp != bprm->filename)
1596                 kfree(bprm->interp);
1597         kfree(bprm->fdpath);
1598         kfree(bprm);
1599 }
1600
1601 static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1602 {
1603         struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1604         int retval = -ENOMEM;
1605         if (!bprm)
1606                 goto out;
1607
1608         if (fd == AT_FDCWD || filename->name[0] == '/') {
1609                 bprm->filename = filename->name;
1610         } else {
1611                 if (filename->name[0] == '\0')
1612                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1613                 else
1614                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1615                                                   fd, filename->name);
1616                 if (!bprm->fdpath)
1617                         goto out_free;
1618
1619                 bprm->filename = bprm->fdpath;
1620         }
1621         bprm->interp = bprm->filename;
1622
1623         retval = bprm_mm_init(bprm);
1624         if (retval)
1625                 goto out_free;
1626         return bprm;
1627
1628 out_free:
1629         free_bprm(bprm);
1630 out:
1631         return ERR_PTR(retval);
1632 }
1633
1634 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1635 {
1636         /* If a binfmt changed the interp, free it first. */
1637         if (bprm->interp != bprm->filename)
1638                 kfree(bprm->interp);
1639         bprm->interp = kstrdup(interp, GFP_KERNEL);
1640         if (!bprm->interp)
1641                 return -ENOMEM;
1642         return 0;
1643 }
1644 EXPORT_SYMBOL(bprm_change_interp);
1645
1646 /*
1647  * determine how safe it is to execute the proposed program
1648  * - the caller must hold ->cred_guard_mutex to protect against
1649  *   PTRACE_ATTACH or seccomp thread-sync
1650  */
1651 static void check_unsafe_exec(struct linux_binprm *bprm)
1652 {
1653         struct task_struct *p = current, *t;
1654         unsigned n_fs;
1655
1656         if (p->ptrace)
1657                 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1658
1659         /*
1660          * This isn't strictly necessary, but it makes it harder for LSMs to
1661          * mess up.
1662          */
1663         if (task_no_new_privs(current))
1664                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1665
1666         t = p;
1667         n_fs = 1;
1668         spin_lock(&p->fs->lock);
1669         rcu_read_lock();
1670         while_each_thread(p, t) {
1671                 if (t->fs == p->fs)
1672                         n_fs++;
1673         }
1674         rcu_read_unlock();
1675
1676         if (p->fs->users > n_fs)
1677                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1678         else
1679                 p->fs->in_exec = 1;
1680         spin_unlock(&p->fs->lock);
1681 }
1682
1683 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1684 {
1685         /* Handle suid and sgid on files */
1686         struct inode *inode;
1687         unsigned int mode;
1688         kuid_t uid;
1689         kgid_t gid;
1690
1691         if (!mnt_may_suid(file->f_path.mnt))
1692                 return;
1693
1694         if (task_no_new_privs(current))
1695                 return;
1696
1697         inode = file->f_path.dentry->d_inode;
1698         mode = READ_ONCE(inode->i_mode);
1699         if (!(mode & (S_ISUID|S_ISGID)))
1700                 return;
1701
1702         /* Be careful if suid/sgid is set */
1703         inode_lock(inode);
1704
1705         /* reload atomically mode/uid/gid now that lock held */
1706         mode = inode->i_mode;
1707         uid = inode->i_uid;
1708         gid = inode->i_gid;
1709         inode_unlock(inode);
1710
1711         /* We ignore suid/sgid if there are no mappings for them in the ns */
1712         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1713                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1714                 return;
1715
1716         if (mode & S_ISUID) {
1717                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1718                 bprm->cred->euid = uid;
1719         }
1720
1721         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1722                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1723                 bprm->cred->egid = gid;
1724         }
1725 }
1726
1727 /*
1728  * Compute brpm->cred based upon the final binary.
1729  */
1730 static int bprm_creds_from_file(struct linux_binprm *bprm)
1731 {
1732         /* Compute creds based on which file? */
1733         struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1734
1735         bprm_fill_uid(bprm, file);
1736         return security_bprm_creds_from_file(bprm, file);
1737 }
1738
1739 /*
1740  * Fill the binprm structure from the inode.
1741  * Read the first BINPRM_BUF_SIZE bytes
1742  *
1743  * This may be called multiple times for binary chains (scripts for example).
1744  */
1745 static int prepare_binprm(struct linux_binprm *bprm)
1746 {
1747         loff_t pos = 0;
1748
1749         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1750         return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1751 }
1752
1753 /*
1754  * Arguments are '\0' separated strings found at the location bprm->p
1755  * points to; chop off the first by relocating brpm->p to right after
1756  * the first '\0' encountered.
1757  */
1758 int remove_arg_zero(struct linux_binprm *bprm)
1759 {
1760         int ret = 0;
1761         unsigned long offset;
1762         char *kaddr;
1763         struct page *page;
1764
1765         if (!bprm->argc)
1766                 return 0;
1767
1768         do {
1769                 offset = bprm->p & ~PAGE_MASK;
1770                 page = get_arg_page(bprm, bprm->p, 0);
1771                 if (!page) {
1772                         ret = -EFAULT;
1773                         goto out;
1774                 }
1775                 kaddr = kmap_atomic(page);
1776
1777                 for (; offset < PAGE_SIZE && kaddr[offset];
1778                                 offset++, bprm->p++)
1779                         ;
1780
1781                 kunmap_atomic(kaddr);
1782                 put_arg_page(page);
1783         } while (offset == PAGE_SIZE);
1784
1785         bprm->p++;
1786         bprm->argc--;
1787         ret = 0;
1788
1789 out:
1790         return ret;
1791 }
1792 EXPORT_SYMBOL(remove_arg_zero);
1793
1794 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1795 /*
1796  * cycle the list of binary formats handler, until one recognizes the image
1797  */
1798 static int search_binary_handler(struct linux_binprm *bprm)
1799 {
1800         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1801         struct linux_binfmt *fmt;
1802         int retval;
1803
1804         retval = prepare_binprm(bprm);
1805         if (retval < 0)
1806                 return retval;
1807
1808         retval = security_bprm_check(bprm);
1809         if (retval)
1810                 return retval;
1811
1812         retval = -ENOENT;
1813  retry:
1814         read_lock(&binfmt_lock);
1815         list_for_each_entry(fmt, &formats, lh) {
1816                 if (!try_module_get(fmt->module))
1817                         continue;
1818                 read_unlock(&binfmt_lock);
1819
1820                 retval = fmt->load_binary(bprm);
1821
1822                 read_lock(&binfmt_lock);
1823                 put_binfmt(fmt);
1824                 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1825                         read_unlock(&binfmt_lock);
1826                         return retval;
1827                 }
1828         }
1829         read_unlock(&binfmt_lock);
1830
1831         if (need_retry) {
1832                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1833                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1834                         return retval;
1835                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1836                         return retval;
1837                 need_retry = false;
1838                 goto retry;
1839         }
1840
1841         return retval;
1842 }
1843
1844 static int exec_binprm(struct linux_binprm *bprm)
1845 {
1846         pid_t old_pid, old_vpid;
1847         int ret, depth;
1848
1849         /* Need to fetch pid before load_binary changes it */
1850         old_pid = current->pid;
1851         rcu_read_lock();
1852         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1853         rcu_read_unlock();
1854
1855         /* This allows 4 levels of binfmt rewrites before failing hard. */
1856         for (depth = 0;; depth++) {
1857                 struct file *exec;
1858                 if (depth > 5)
1859                         return -ELOOP;
1860
1861                 ret = search_binary_handler(bprm);
1862                 if (ret < 0)
1863                         return ret;
1864                 if (!bprm->interpreter)
1865                         break;
1866
1867                 exec = bprm->file;
1868                 bprm->file = bprm->interpreter;
1869                 bprm->interpreter = NULL;
1870
1871                 allow_write_access(exec);
1872                 if (unlikely(bprm->have_execfd)) {
1873                         if (bprm->executable) {
1874                                 fput(exec);
1875                                 return -ENOEXEC;
1876                         }
1877                         bprm->executable = exec;
1878                 } else
1879                         fput(exec);
1880         }
1881
1882         audit_bprm(bprm);
1883         trace_sched_process_exec(current, old_pid, bprm);
1884         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1885         proc_exec_connector(current);
1886         return 0;
1887 }
1888
1889 /*
1890  * sys_execve() executes a new program.
1891  */
1892 static int bprm_execve(struct linux_binprm *bprm,
1893                        int fd, struct filename *filename, int flags)
1894 {
1895         struct file *file;
1896         struct files_struct *displaced;
1897         int retval;
1898
1899         retval = unshare_files(&displaced);
1900         if (retval)
1901                 return retval;
1902
1903         retval = prepare_bprm_creds(bprm);
1904         if (retval)
1905                 goto out_files;
1906
1907         check_unsafe_exec(bprm);
1908         current->in_execve = 1;
1909
1910         file = do_open_execat(fd, filename, flags);
1911         retval = PTR_ERR(file);
1912         if (IS_ERR(file))
1913                 goto out_unmark;
1914
1915         sched_exec();
1916
1917         bprm->file = file;
1918         /*
1919          * Record that a name derived from an O_CLOEXEC fd will be
1920          * inaccessible after exec. Relies on having exclusive access to
1921          * current->files (due to unshare_files above).
1922          */
1923         if (bprm->fdpath &&
1924             close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1925                 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1926
1927         /* Set the unchanging part of bprm->cred */
1928         retval = security_bprm_creds_for_exec(bprm);
1929         if (retval)
1930                 goto out;
1931
1932         retval = exec_binprm(bprm);
1933         if (retval < 0)
1934                 goto out;
1935
1936         /* execve succeeded */
1937         current->fs->in_exec = 0;
1938         current->in_execve = 0;
1939         rseq_execve(current);
1940         acct_update_integrals(current);
1941         task_numa_free(current, false);
1942         if (displaced)
1943                 put_files_struct(displaced);
1944         return retval;
1945
1946 out:
1947         /*
1948          * If past the point of no return ensure the the code never
1949          * returns to the userspace process.  Use an existing fatal
1950          * signal if present otherwise terminate the process with
1951          * SIGSEGV.
1952          */
1953         if (bprm->point_of_no_return && !fatal_signal_pending(current))
1954                 force_sigsegv(SIGSEGV);
1955
1956 out_unmark:
1957         current->fs->in_exec = 0;
1958         current->in_execve = 0;
1959
1960 out_files:
1961         if (displaced)
1962                 reset_files_struct(displaced);
1963
1964         return retval;
1965 }
1966
1967 static int do_execveat_common(int fd, struct filename *filename,
1968                               struct user_arg_ptr argv,
1969                               struct user_arg_ptr envp,
1970                               int flags)
1971 {
1972         struct linux_binprm *bprm;
1973         int retval;
1974
1975         if (IS_ERR(filename))
1976                 return PTR_ERR(filename);
1977
1978         /*
1979          * We move the actual failure in case of RLIMIT_NPROC excess from
1980          * set*uid() to execve() because too many poorly written programs
1981          * don't check setuid() return code.  Here we additionally recheck
1982          * whether NPROC limit is still exceeded.
1983          */
1984         if ((current->flags & PF_NPROC_EXCEEDED) &&
1985             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1986                 retval = -EAGAIN;
1987                 goto out_ret;
1988         }
1989
1990         /* We're below the limit (still or again), so we don't want to make
1991          * further execve() calls fail. */
1992         current->flags &= ~PF_NPROC_EXCEEDED;
1993
1994         bprm = alloc_bprm(fd, filename);
1995         if (IS_ERR(bprm)) {
1996                 retval = PTR_ERR(bprm);
1997                 goto out_ret;
1998         }
1999
2000         retval = count(argv, MAX_ARG_STRINGS);
2001         if (retval < 0)
2002                 goto out_free;
2003         bprm->argc = retval;
2004
2005         retval = count(envp, MAX_ARG_STRINGS);
2006         if (retval < 0)
2007                 goto out_free;
2008         bprm->envc = retval;
2009
2010         retval = bprm_stack_limits(bprm);
2011         if (retval < 0)
2012                 goto out_free;
2013
2014         retval = copy_string_kernel(bprm->filename, bprm);
2015         if (retval < 0)
2016                 goto out_free;
2017         bprm->exec = bprm->p;
2018
2019         retval = copy_strings(bprm->envc, envp, bprm);
2020         if (retval < 0)
2021                 goto out_free;
2022
2023         retval = copy_strings(bprm->argc, argv, bprm);
2024         if (retval < 0)
2025                 goto out_free;
2026
2027         retval = bprm_execve(bprm, fd, filename, flags);
2028 out_free:
2029         free_bprm(bprm);
2030
2031 out_ret:
2032         putname(filename);
2033         return retval;
2034 }
2035
2036 int kernel_execve(const char *kernel_filename,
2037                   const char *const *argv, const char *const *envp)
2038 {
2039         struct filename *filename;
2040         struct linux_binprm *bprm;
2041         int fd = AT_FDCWD;
2042         int retval;
2043
2044         filename = getname_kernel(kernel_filename);
2045         if (IS_ERR(filename))
2046                 return PTR_ERR(filename);
2047
2048         bprm = alloc_bprm(fd, filename);
2049         if (IS_ERR(bprm)) {
2050                 retval = PTR_ERR(bprm);
2051                 goto out_ret;
2052         }
2053
2054         retval = count_strings_kernel(argv);
2055         if (retval < 0)
2056                 goto out_free;
2057         bprm->argc = retval;
2058
2059         retval = count_strings_kernel(envp);
2060         if (retval < 0)
2061                 goto out_free;
2062         bprm->envc = retval;
2063
2064         retval = bprm_stack_limits(bprm);
2065         if (retval < 0)
2066                 goto out_free;
2067
2068         retval = copy_string_kernel(bprm->filename, bprm);
2069         if (retval < 0)
2070                 goto out_free;
2071         bprm->exec = bprm->p;
2072
2073         retval = copy_strings_kernel(bprm->envc, envp, bprm);
2074         if (retval < 0)
2075                 goto out_free;
2076
2077         retval = copy_strings_kernel(bprm->argc, argv, bprm);
2078         if (retval < 0)
2079                 goto out_free;
2080
2081         retval = bprm_execve(bprm, fd, filename, 0);
2082 out_free:
2083         free_bprm(bprm);
2084 out_ret:
2085         putname(filename);
2086         return retval;
2087 }
2088
2089 static int do_execve(struct filename *filename,
2090         const char __user *const __user *__argv,
2091         const char __user *const __user *__envp)
2092 {
2093         struct user_arg_ptr argv = { .ptr.native = __argv };
2094         struct user_arg_ptr envp = { .ptr.native = __envp };
2095         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2096 }
2097
2098 static int do_execveat(int fd, struct filename *filename,
2099                 const char __user *const __user *__argv,
2100                 const char __user *const __user *__envp,
2101                 int flags)
2102 {
2103         struct user_arg_ptr argv = { .ptr.native = __argv };
2104         struct user_arg_ptr envp = { .ptr.native = __envp };
2105
2106         return do_execveat_common(fd, filename, argv, envp, flags);
2107 }
2108
2109 #ifdef CONFIG_COMPAT
2110 static int compat_do_execve(struct filename *filename,
2111         const compat_uptr_t __user *__argv,
2112         const compat_uptr_t __user *__envp)
2113 {
2114         struct user_arg_ptr argv = {
2115                 .is_compat = true,
2116                 .ptr.compat = __argv,
2117         };
2118         struct user_arg_ptr envp = {
2119                 .is_compat = true,
2120                 .ptr.compat = __envp,
2121         };
2122         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2123 }
2124
2125 static int compat_do_execveat(int fd, struct filename *filename,
2126                               const compat_uptr_t __user *__argv,
2127                               const compat_uptr_t __user *__envp,
2128                               int flags)
2129 {
2130         struct user_arg_ptr argv = {
2131                 .is_compat = true,
2132                 .ptr.compat = __argv,
2133         };
2134         struct user_arg_ptr envp = {
2135                 .is_compat = true,
2136                 .ptr.compat = __envp,
2137         };
2138         return do_execveat_common(fd, filename, argv, envp, flags);
2139 }
2140 #endif
2141
2142 void set_binfmt(struct linux_binfmt *new)
2143 {
2144         struct mm_struct *mm = current->mm;
2145
2146         if (mm->binfmt)
2147                 module_put(mm->binfmt->module);
2148
2149         mm->binfmt = new;
2150         if (new)
2151                 __module_get(new->module);
2152 }
2153 EXPORT_SYMBOL(set_binfmt);
2154
2155 /*
2156  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2157  */
2158 void set_dumpable(struct mm_struct *mm, int value)
2159 {
2160         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2161                 return;
2162
2163         set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2164 }
2165
2166 SYSCALL_DEFINE3(execve,
2167                 const char __user *, filename,
2168                 const char __user *const __user *, argv,
2169                 const char __user *const __user *, envp)
2170 {
2171         return do_execve(getname(filename), argv, envp);
2172 }
2173
2174 SYSCALL_DEFINE5(execveat,
2175                 int, fd, const char __user *, filename,
2176                 const char __user *const __user *, argv,
2177                 const char __user *const __user *, envp,
2178                 int, flags)
2179 {
2180         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2181
2182         return do_execveat(fd,
2183                            getname_flags(filename, lookup_flags, NULL),
2184                            argv, envp, flags);
2185 }
2186
2187 #ifdef CONFIG_COMPAT
2188 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2189         const compat_uptr_t __user *, argv,
2190         const compat_uptr_t __user *, envp)
2191 {
2192         return compat_do_execve(getname(filename), argv, envp);
2193 }
2194
2195 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2196                        const char __user *, filename,
2197                        const compat_uptr_t __user *, argv,
2198                        const compat_uptr_t __user *, envp,
2199                        int,  flags)
2200 {
2201         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2202
2203         return compat_do_execveat(fd,
2204                                   getname_flags(filename, lookup_flags, NULL),
2205                                   argv, envp, flags);
2206 }
2207 #endif